aifollow.news 搜索
返回 IT之家 科技新闻
IT之家 科技新闻· · 原发布时间

宇树科技王兴兴披露机器人测试进展与“宇树预演”开发计划

自动核验发布 · 本文由系统生成并完成证据核验,未经人工审稿。

AI 辅助摘要

王兴兴在演讲中称,宇树测试显示,机器人在零训练条件下已能由模型控制完成画画、清洗画笔和下围棋等任务。宇树还在开发“宇树预演”,计划让 AI 模型在仿真与真实环境中自行训练、测试并获取反馈;他同时指出,机器人与真实物体交互及模型对齐仍是待解决的问题。

正文 · AI 翻译

Thanks to IT Home reader 不一样的体验 for the tip!

IT Home reported on October 22 that, according to Sina Technology today, Unitree Robotics founder Wang Xingxing gave a speech at the 2026 World Physical AI Conference, reviewing the company's progress over the past year and sharing his latest assessment of trends in physical AI.

Wang Xingxing revealed that the Unitree G1 is currently the world's best-selling humanoid robot, with approximately 10,000 to 20,000 units produced to date. Discussing the impact of GPT-6's release on robotics, Wang said Unitree's tests showed that, with no training at all, the model could already directly control a robot to perform tasks such as painting, cleaning a paintbrush and playing Go. The robot can change colors and clean its paintbrush on its own, and “the precision of its moves is very high.”

He predicted that major AI model companies would continue to pay increasing attention to machine intelligence and the robotics industry in the coming years. “If things move quickly, the industry could see more progress in 1 to 2 years; if more slowly, in 2 to 3 years.”

Wang Xingxing said he hopes that, in the future, robots will be able to complete about 80% of tasks using only spoken and language instructions. “It might be possible as soon as next year; a more cautious estimate would be the year after next.” He noted that the main issues still not adequately resolved are how robots interact with real objects and how to align AI model inputs and outputs with physical robots. A large part of this, he said, involves engineering challenges.

IT Home learned from the original report that Wang Xingxing also revealed that Unitree is developing “Unitree Preview,” a self-evolving system for physical AI robots that would let AI models collect research papers, write code, train in simulated environments and then run tests in real-world environments. The AI models would score the test results and provide feedback, creating a self-sustaining cycle.

At the end of his speech, Wang Xingxing admitted, “I'm really afraid of wasting every day,” and said, “Our generation truly has a chance to achieve superintelligence or AGI, and also a chance to have humanoid robots take productivity to the next level.”

发现内容有误?提交纠错